Database Reference
In-Depth Information
Gutierrez:
What problems did you apply the system to?
Radinsky:
At this point, I was working with Eric Horvitz who has an MD and
is the head of the Redmond Lab at Microsoft Research. He is extremely inter-
ested in disease prediction and algorithms. Once the system was taking into
account correlations, we decided to see if the system could predict cholera
outbreaks. We wrote some code and made it so that we could tell the system
“cholera outbreak” and the system would give me the probability every day
of that event happening in different places in the world. When we looked at
historical data, it was right 80-something percent of the time. And more than
that, in 2012, we predicted the first cholera outbreak in Cuba in 130 years.
What happens is that cholera is a waterborne disease, so you wouldn't be sur-
prised that the system found out that it tends to happen after floods. But the
pattern that the system found was that if two years before those floods you
have a drought, the probability of those floods leading to a cholera outbreak is
much, much higher. And not only that—it usually tends to happen in countries
with low GDP and low concentrations of water. Why the low concentration
of water? Eric told me that cholera is treated very easily by having clean water.
Clean water drops the mortality rate from 50% to less than 1%. So people
who had access to clean water just don't have outbreaks, which is why being
a low-GDP country matters.
Having achieved this breakthrough in cholera, we then started looking at pre-
dicting riots. For example, the system predicted the latest riots in Sudan.
What we found out is that if you have a basic product in the country and the
price of this product starts going up, then you're going to have student riots.
Then, if in those student riots, a policeman kills one of the students, this will
lead to much bigger riots that can even affect the government. This was what
happened in Egypt with the bread prices. The system inferred from Egypt that
the same thing was going to happen in Sudan when the gas price started to go
up. In Sudan, the government had subsidized the price of gas for many years.
The events in Sudan unfolded just like the system predicted—in December
there were student riots, then a policeman killed one of the students, which
in turn caused huge riots.
For me, it was exciting that the system predicted the cholera outbreak and
the riots; it was the first time I'd seen a system do that. The system can change
the way we see the world, because it can find patterns that we've never seen
before. This is the dream of every scientist—a better understanding of the
world we live in. This is a really great tool for decision makers, who are mak-
ing decisions in the dark that affect the lives of all of us. I just want to give
them the scientific tools to help them make better decisions, because we
already have tons of data. We can help them so much. We've recently started
working with the Bill Gates Foundation to build a much more granular cholera
predictor to try to predict cholera outbreaks all over the world.
